Development of an agent model based on naturalistic data to simulate rider’s braking maneuvers

Mobility has been changed in the last decade and it is supposed to change in the next years. New means of transportation and new road users will be present on public roads. To improve road safety, road users’ behavior has to be studied as well as their interactions. For this reason, since 1950s researchers tried to model how users move in traffi c. Although many papers have deeply investigated car drivers’ behavior and many diff erent models have been developed, other road users’ behavior has not been so well studied and modeled. As a matter of fact, in traffi c simulators, motorcyclists’ models are usually derived from car drivers’ ones, despite their behaviors are diff erent (lane-based vs. non-lane-based). Each user should be simulated with specifi c models to make simulations more realistic in such a developing and complex scenario, as the current one, where heterogeneous traffi c fl ows are increasingly relevant because of the coexistence of vehicles (both conventional and with diff erent levels of automation), PTW, bicycles (both conventional and electric), e-scooters, mono-wheel, etc. For this reason in this thesis a new kind of behavioral model is proposed to simulate PTWs. The main goal was to obtain a model (suggesting new approach) specifi c for riders. The proposed architecture includes AI algorithms to make the model able to adapt its responses to diff erent conditions without rigid constrains that can be found in classical mathematical behavioral models. For this reason Naturalistic data has been analyzed, to identify the main factors that drive riders’ decision-making process and their maneuvers execution. The model was developed according to the agent-based modeling approach, with a specifi c focus on micro-simulations. The whole development of a new model goes beyond the eff ort of a PhD, this thesis focused on the development of the model architecture and of the braking maneuver, being the most frequent action in safety-critical conditions. Since driving behavior can be modeled at diff erent levels, the presented model focused on Tactical and Operational ones. The proposed study is divided in four parts. First a naturalistic riding database was analyzed; then, according to the obtained results, the architecture of the model was defi ned and developed. The performance of each part and the whole model was evaluated, and fi nally the model was applied to a case study and integrated in a microscopic traffi c simulation tool. The study showed the feasibility of a new approach that can reproduce riders’ maneuvers with high level of realism.